The application of machine learning to rural population migration research

Many rural areas experience population stagnation or decline from out-migration with corresponding economic downturns. This is the case for the Northern Territory in Australia, a vast and sparsely populated jurisdiction. Its government has long sought to encourage stronger population growth but its population is young and highly transient, leading to high staff turn-overs and challenges for industries and government to attract families and skilled workers. Place-based factors such as job opportunities, access to essential services or environmental amenities influence satisfaction and migration decisions. The aim of this study was to understand why people might stay or move away through analysing responses to two open-text questions on the best and worst aspect of living in the Northern Territory. Over 3500 valid responses were analysed using machine learning-based unsupervised topic modelling which uncovered latent clusters. Forty-four percent of positive aspects were clustered into lifestyle factors, while negative aspects clustered around high living costs and crime. Some aspects, such as the weather and distance to other places were discussed as both positive and negative aspects. Topics discussed by respondents could be directly related to their intention to leave the Northern Territory, and also to specific individual's demographic characteristics providing insights for policies focused on attracting and retaining population. The use of unsupervised text mining in population research is rare and this study verifies its use to deliver objective and nuanced results generated from a large qualitative data set.